AI robotics startup Astrocade emerges from stealth with $71M seed and live robot interaction platform
Key Points
- Astrocade launches from stealth with a $71M seed round led by Ribbit Capital and a live platform letting users control physical robots through AI foundation models.
- The 26-person company bets on software-first robotics, aiming to port its model stack to new hardware embodiments within a week rather than building proprietary robots.
- Founder Jonathan Jacobi argues specialized robotic foundation models will outcompete general LLMs on cost and reliability, dismissing frontier models as impractical for real-world deployment.
Summary
Read full transcript →Astrocade emerges from stealth with $71M seed
Astrocade, a 26-person AI robotics research company, announced a $71M seed round led by Ribbit Capital and simultaneously launched a live online platform letting users interact with real robots via its foundation models.
The launch platform gives the concept immediate texture. Users can browse to a website and control physical robots in real time across several modes: a painter robot that takes a word association and plays Pictionary, a sword-fighting robot that responds to a human opponent via AI, and a robot that responds to natural language commands and performs pantomimes. Jacobi describes it as a deliberate attempt to learn how people want to interact with robots and feed that learning back into research.
“Today we launched the name out of stealth, raising a $71,000,000 seed round. We launched the first AI robotic experience online so people can just browse to a website, interact with real AI powered robots via our models and interfaces. We're a team of 26 people.”
Software, not hardware
Astrocade positions itself as a software-first company. The goal is to generalize its model stack across robot embodiments so that when a new hardware manufacturer ships a robot, Astrocade can have its full technology running on that platform within a week. Jacobi argues this decoupling is viable because there are already strong hardware vendors globally, and the gap is in the "brain and interface" layer.
Foundation models vs. LLM control
On the question of whether general-purpose LLMs are eating into specialized robotics models, Jacobi is skeptical. He acknowledges demos of frontier models controlling robotic arms but argues they don't generalize well and don't meet the dexterity or cost requirements of real deployments. His analogy: you could theoretically use an LLM to generate ASCII art, but image models exist because they're the right tool. Dedicated robotic foundation models will similarly win on cost and reliability.
Commercial timeline
Astrocade is already working with partners across logistics, entertainment, and healthcare, though Jacobi stops short of specific revenue figures or timelines. He says the company is seeing value from partners faster than expected, and is being deliberate about scaling to avoid "product debt" from locking into the wrong direction early.
Before founding Astrocade, Jacobi worked at a company acquired by Waze, which was subsequently acquired by Google. He has a long working relationship with Waze's CEO going back to when he joined Microsoft in Israel at 17.
Every deal, every interview. 5 minutes.
TBPN Digest delivers summaries of the latest fundraises, interviews and tech news from TBPN, every weekday.